ECG-ArrestNetIHCA Prediction Platform
Department of Cardiology · Central South University
RESEARCH-USE ECG ANALYSIS PLATFORM

ECG-ArrestNet: In-Hospital Cardiac Arrest Prediction from 12-Lead ECG

Department of Cardiology, Second Xiangya Hospital, Central South University, Changsha, China
This research prototype demonstrates the inference workflow of ECG-ArrestNet, a deep learning model developed to estimate the risk of in-hospital cardiac arrest (IHCA) within 24 hours after index ECG acquisition from raw 12-lead ECG signals. Users may upload a recording, run the inference pipeline, and obtain a model-estimated IHCA probability with feature-level attribution. The model integrates a multi-scale 1D-CNN backbone, lead-aware cross-attention, a BiLSTM temporal encoder, and gated fusion of 52 ECG-derived features with deep-learning representations.
Uploaded ECG data are processed under strict privacy protection and never stored or shared.
01 — Prediction Workspace

Analyze a 12-lead ECG

Upload a raw 12-lead ECG recording (GuiTu XML). ECG-ArrestNet extracts electrophysiologic features, runs the full inference pipeline, and returns a model-estimated IHCA probability with explainable contributions.

AECG Input
Upload a signal
12-LEAD ECG · 500 HzNo signal loaded
Upload a GuiTu XML signal, then run the prediction.
BInference & Results
Idle
✓
Stage 1
Multi-scale 1D-CNN
k=3/7/15/31 per lead
✓
Stage 2
Cross-Attention
8-head lead-aware
✓
Stage 3
BiLSTM Encoder
128 units, temporal
✓
Stage 4
Gated Fusion
ECG-derived features + deep representation
Run a prediction to view the IHCA risk score, feature contributions, and lead attention.
CSession Log
  • No analyses yet.
02 - About

Research prototype

ECG-ArrestNet is a research prototype developed to demonstrate an ECG-based deep learning workflow for estimating short-term IHCA risk after index ECG acquisition.

HHospital

Second Xiangya Hospital of Central South University — a tertiary referral center providing the 12-lead ECG data for ECG-ArrestNet development and validation.

AArchitecture

Multi-scale 1D-CNN backbone, 8-head lead-aware cross-attention, BiLSTM temporal encoder, and gated fusion of 52 ECG-derived features with deep-learning representations.

PPrivacy

Uploaded ECG data are processed under strict data-protection safeguards and are never stored, shared, or used for any other purpose.

How to use this platform

Upload a 12-lead ECG file (GuiTu XML resting-ECG export), then run the prediction. The model uses the ECG signal alone — no demographic or clinical inputs. This in-browser demonstration reproduces the ECG-ArrestNet inference workflow for research illustration; the output includes a model-estimated IHCA probability, classification relative to the model threshold, feature-level contributions, and a lead-attention visualization.

03 — Notice

Important notice

Research prototype — not for clinical use

This platform is a research demonstration of the ECG-ArrestNet inference workflow. It is not a medical device and must not be used for diagnosis, triage, monitoring, or treatment decisions. As stated in the study, the model was evaluated retrospectively; prospective validation — ideally in a randomized controlled trial — is required before any clinical adoption. Uploaded signals are processed locally in the browser and are never transmitted or stored.

!Retrospective validation

Reported performance derives from retrospective case-control cohorts and may differ in prospective, real-time settings due to data quality, workflow, and population shifts.

!Prevalence calibration

Study cohorts used an enriched case-control design (~25% positive rate), not real-world IHCA prevalence. PPV and NPV are prevalence-dependent and can be recalibrated for any target prevalence using Bayes' theorem.